A framework for online dynamic channel weighting is developed for the task of EEG-based neonatal seizure detection. The channel weights are computed on-the-fly by combining the up-to-now patient-specific history and the clinically-derived prior channel importance. These estimated time-varying weights are introduced within a Bayesian probabilistic framework to provide a channel-specific and thus patient-adaptive seizure classification scheme. Validation results on one of the largest clinical datasets of neonatal seizures confirm the utility of the proposed channel weighting for the SVM-based detector recently developed by this research group. Exploiting the channel weighting, the precision-recall area can be drastically increased (up to 25%) for the most difficult patients, with the average increase from 81.0% to 84.42%. It is also shown that the increase in performance with channel weighting is proportional to the time the patient is observed.